From: Philip Goetz (philgoetz@gmail.com)
Date: Mon Jul 31 2006 - 16:31:55 MDT
On 7/31/06, Eliezer S. Yudkowsky <sentience@pobox.com> wrote:
> Goetz wrote:
> >
> > 1. Establishment of the world's most massive computing centers - this
> > is odd, given that Google already has the best response time and most
> > reliable websites in the world, and the amount of centralized
> > computing power they are building does not seem necessary for indexing
> > the web using anything like present means. Although I don't really
> > know.
> >
> > 2. Scanning in and converting to computer-searchable files as many
> > books as they can.
> >
> > When an AI is developed, Google will have the most computer power to
> > run it on, and the largest collection of data to feed it with.
>
> Size matters not. Have I taught you nothing?
I think you may be right about computing power. The notion being, I
suppose, that once you get to better-than-human, you (or your AI, at
least :) have won. However, it is likely that algorithms will be
publicly known that will not produce an AI on one computer, but will
produce it on a network of a million servers.
Data, in any case, matters a lot. Although the difference between
having the internet, and the internet plus a lot of scanned in books,
might not be crucial. I think having a lot of journal articles might
be more important. The web contains, for the most part, only basic
information on any given discipline. Textbooks contain another level
of detail. Journal articles contain a yet-higher level of detail.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT